blob: e8918ffaf96d4a0a2dacf75838b5d8a89e5e8ca3 [file] [log] [blame]
Vivek Goyal31e4c282009-12-03 12:59:42 -05001/*
2 * Common Block IO controller cgroup interface
3 *
4 * Based on ideas and code from CFQ, CFS and BFQ:
5 * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
6 *
7 * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it>
8 * Paolo Valente <paolo.valente@unimore.it>
9 *
10 * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com>
11 * Nauman Rafique <nauman@google.com>
12 */
13#include <linux/ioprio.h>
Vivek Goyal22084192009-12-03 12:59:49 -050014#include <linux/kdev_t.h>
Vivek Goyal9d6a9862009-12-04 10:36:41 -050015#include <linux/module.h>
Stephen Rothwellaccee782009-12-07 19:29:39 +110016#include <linux/err.h>
Divyesh Shah91952912010-04-01 15:01:41 -070017#include <linux/blkdev.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090018#include <linux/slab.h>
Gui Jianfeng34d0f172010-04-13 16:05:49 +080019#include <linux/genhd.h>
Tejun Heo72e06c22012-03-05 13:15:00 -080020#include <linux/delay.h>
Tejun Heo9a9e8a22012-03-19 15:10:56 -070021#include <linux/atomic.h>
Tejun Heo72e06c22012-03-05 13:15:00 -080022#include "blk-cgroup.h"
Tejun Heo5efd6112012-03-05 13:15:12 -080023#include "blk.h"
Vivek Goyal3e252062009-12-04 10:36:42 -050024
Divyesh Shah84c124d2010-04-09 08:31:19 +020025#define MAX_KEY_LEN 100
26
Tejun Heobc0d6502012-04-13 13:11:26 -070027static DEFINE_MUTEX(blkcg_pol_mutex);
Tejun Heo923adde2012-03-05 13:15:13 -080028
Tejun Heoe71357e2013-01-09 08:05:10 -080029struct blkcg blkcg_root = { .cfq_weight = 2 * CFQ_WEIGHT_DEFAULT,
30 .cfq_leaf_weight = 2 * CFQ_WEIGHT_DEFAULT, };
Tejun Heo3c798392012-04-16 13:57:25 -070031EXPORT_SYMBOL_GPL(blkcg_root);
Vivek Goyal9d6a9862009-12-04 10:36:41 -050032
Tejun Heo3c798392012-04-16 13:57:25 -070033static struct blkcg_policy *blkcg_policy[BLKCG_MAX_POLS];
Tejun Heo035d10b2012-03-05 13:15:04 -080034
Tejun Heo16b3de62013-01-09 08:05:12 -080035static struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg,
36 struct request_queue *q, bool update_hint);
37
38/**
39 * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
40 * @d_blkg: loop cursor pointing to the current descendant
41 * @pos_cgrp: used for iteration
42 * @p_blkg: target blkg to walk descendants of
43 *
44 * Walk @c_blkg through the descendants of @p_blkg. Must be used with RCU
45 * read locked. If called under either blkcg or queue lock, the iteration
46 * is guaranteed to include all and only online blkgs. The caller may
47 * update @pos_cgrp by calling cgroup_rightmost_descendant() to skip
48 * subtree.
49 */
50#define blkg_for_each_descendant_pre(d_blkg, pos_cgrp, p_blkg) \
51 cgroup_for_each_descendant_pre((pos_cgrp), (p_blkg)->blkcg->css.cgroup) \
52 if (((d_blkg) = __blkg_lookup(cgroup_to_blkcg(pos_cgrp), \
53 (p_blkg)->q, false)))
54
Tejun Heoa2b16932012-04-13 13:11:33 -070055static bool blkcg_policy_enabled(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -070056 const struct blkcg_policy *pol)
Tejun Heoa2b16932012-04-13 13:11:33 -070057{
58 return pol && test_bit(pol->plid, q->blkcg_pols);
59}
60
Tejun Heo03814112012-03-05 13:15:14 -080061/**
62 * blkg_free - free a blkg
63 * @blkg: blkg to free
64 *
65 * Free @blkg which may be partially allocated.
66 */
Tejun Heo3c798392012-04-16 13:57:25 -070067static void blkg_free(struct blkcg_gq *blkg)
Tejun Heo03814112012-03-05 13:15:14 -080068{
Tejun Heoe8989fa2012-03-05 13:15:20 -080069 int i;
Tejun Heo549d3aa2012-03-05 13:15:16 -080070
71 if (!blkg)
72 return;
73
Tejun Heo8bd435b2012-04-13 13:11:28 -070074 for (i = 0; i < BLKCG_MAX_POLS; i++) {
Tejun Heo3c798392012-04-16 13:57:25 -070075 struct blkcg_policy *pol = blkcg_policy[i];
Tejun Heoe8989fa2012-03-05 13:15:20 -080076 struct blkg_policy_data *pd = blkg->pd[i];
77
Tejun Heo9ade5ea2012-04-01 14:38:44 -070078 if (!pd)
79 continue;
80
Tejun Heof9fcc2d2012-04-16 13:57:27 -070081 if (pol && pol->pd_exit_fn)
82 pol->pd_exit_fn(blkg);
Tejun Heo9ade5ea2012-04-01 14:38:44 -070083
Tejun Heo9ade5ea2012-04-01 14:38:44 -070084 kfree(pd);
Tejun Heo03814112012-03-05 13:15:14 -080085 }
Tejun Heoe8989fa2012-03-05 13:15:20 -080086
Tejun Heoa0516612012-06-26 15:05:44 -070087 blk_exit_rl(&blkg->rl);
Tejun Heo549d3aa2012-03-05 13:15:16 -080088 kfree(blkg);
Tejun Heo03814112012-03-05 13:15:14 -080089}
90
91/**
92 * blkg_alloc - allocate a blkg
93 * @blkcg: block cgroup the new blkg is associated with
94 * @q: request_queue the new blkg is associated with
Tejun Heo15974992012-06-04 20:40:52 -070095 * @gfp_mask: allocation mask to use
Tejun Heo03814112012-03-05 13:15:14 -080096 *
Tejun Heoe8989fa2012-03-05 13:15:20 -080097 * Allocate a new blkg assocating @blkcg and @q.
Tejun Heo03814112012-03-05 13:15:14 -080098 */
Tejun Heo15974992012-06-04 20:40:52 -070099static struct blkcg_gq *blkg_alloc(struct blkcg *blkcg, struct request_queue *q,
100 gfp_t gfp_mask)
Tejun Heo03814112012-03-05 13:15:14 -0800101{
Tejun Heo3c798392012-04-16 13:57:25 -0700102 struct blkcg_gq *blkg;
Tejun Heoe8989fa2012-03-05 13:15:20 -0800103 int i;
Tejun Heo03814112012-03-05 13:15:14 -0800104
105 /* alloc and init base part */
Tejun Heo15974992012-06-04 20:40:52 -0700106 blkg = kzalloc_node(sizeof(*blkg), gfp_mask, q->node);
Tejun Heo03814112012-03-05 13:15:14 -0800107 if (!blkg)
108 return NULL;
109
Tejun Heoc875f4d2012-03-05 13:15:22 -0800110 blkg->q = q;
Tejun Heoe8989fa2012-03-05 13:15:20 -0800111 INIT_LIST_HEAD(&blkg->q_node);
Tejun Heo03814112012-03-05 13:15:14 -0800112 blkg->blkcg = blkcg;
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800113 blkg->refcnt = 1;
Tejun Heo03814112012-03-05 13:15:14 -0800114
Tejun Heoa0516612012-06-26 15:05:44 -0700115 /* root blkg uses @q->root_rl, init rl only for !root blkgs */
116 if (blkcg != &blkcg_root) {
117 if (blk_init_rl(&blkg->rl, q, gfp_mask))
118 goto err_free;
119 blkg->rl.blkg = blkg;
120 }
121
Tejun Heo8bd435b2012-04-13 13:11:28 -0700122 for (i = 0; i < BLKCG_MAX_POLS; i++) {
Tejun Heo3c798392012-04-16 13:57:25 -0700123 struct blkcg_policy *pol = blkcg_policy[i];
Tejun Heoe8989fa2012-03-05 13:15:20 -0800124 struct blkg_policy_data *pd;
Tejun Heo03814112012-03-05 13:15:14 -0800125
Tejun Heoa2b16932012-04-13 13:11:33 -0700126 if (!blkcg_policy_enabled(q, pol))
Tejun Heoe8989fa2012-03-05 13:15:20 -0800127 continue;
Tejun Heo549d3aa2012-03-05 13:15:16 -0800128
Tejun Heoe8989fa2012-03-05 13:15:20 -0800129 /* alloc per-policy data and attach it to blkg */
Tejun Heo15974992012-06-04 20:40:52 -0700130 pd = kzalloc_node(pol->pd_size, gfp_mask, q->node);
Tejun Heoa0516612012-06-26 15:05:44 -0700131 if (!pd)
132 goto err_free;
Tejun Heo549d3aa2012-03-05 13:15:16 -0800133
Tejun Heoe8989fa2012-03-05 13:15:20 -0800134 blkg->pd[i] = pd;
135 pd->blkg = blkg;
Tejun Heob276a872013-01-09 08:05:12 -0800136 pd->plid = i;
Tejun Heo03814112012-03-05 13:15:14 -0800137
Tejun Heo9b2ea862012-06-04 15:21:00 +0900138 /* invoke per-policy init */
Tejun Heo356d2e52013-01-09 08:05:10 -0800139 if (pol->pd_init_fn)
Tejun Heof9fcc2d2012-04-16 13:57:27 -0700140 pol->pd_init_fn(blkg);
Tejun Heoe8989fa2012-03-05 13:15:20 -0800141 }
142
Tejun Heo03814112012-03-05 13:15:14 -0800143 return blkg;
Tejun Heoa0516612012-06-26 15:05:44 -0700144
145err_free:
146 blkg_free(blkg);
147 return NULL;
Tejun Heo03814112012-03-05 13:15:14 -0800148}
149
Tejun Heo16b3de62013-01-09 08:05:12 -0800150/**
151 * __blkg_lookup - internal version of blkg_lookup()
152 * @blkcg: blkcg of interest
153 * @q: request_queue of interest
154 * @update_hint: whether to update lookup hint with the result or not
155 *
156 * This is internal version and shouldn't be used by policy
157 * implementations. Looks up blkgs for the @blkcg - @q pair regardless of
158 * @q's bypass state. If @update_hint is %true, the caller should be
159 * holding @q->queue_lock and lookup hint is updated on success.
160 */
Tejun Heo3c798392012-04-16 13:57:25 -0700161static struct blkcg_gq *__blkg_lookup(struct blkcg *blkcg,
Tejun Heo86cde6b2013-01-09 08:05:10 -0800162 struct request_queue *q, bool update_hint)
Tejun Heo80fd9972012-04-13 14:50:53 -0700163{
Tejun Heo3c798392012-04-16 13:57:25 -0700164 struct blkcg_gq *blkg;
Tejun Heo80fd9972012-04-13 14:50:53 -0700165
Tejun Heoa6371202012-04-19 16:29:24 -0700166 blkg = rcu_dereference(blkcg->blkg_hint);
167 if (blkg && blkg->q == q)
168 return blkg;
169
170 /*
Tejun Heo86cde6b2013-01-09 08:05:10 -0800171 * Hint didn't match. Look up from the radix tree. Note that the
172 * hint can only be updated under queue_lock as otherwise @blkg
173 * could have already been removed from blkg_tree. The caller is
174 * responsible for grabbing queue_lock if @update_hint.
Tejun Heoa6371202012-04-19 16:29:24 -0700175 */
176 blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id);
Tejun Heo86cde6b2013-01-09 08:05:10 -0800177 if (blkg && blkg->q == q) {
178 if (update_hint) {
179 lockdep_assert_held(q->queue_lock);
180 rcu_assign_pointer(blkcg->blkg_hint, blkg);
181 }
Tejun Heoa6371202012-04-19 16:29:24 -0700182 return blkg;
Tejun Heo86cde6b2013-01-09 08:05:10 -0800183 }
Tejun Heoa6371202012-04-19 16:29:24 -0700184
Tejun Heo80fd9972012-04-13 14:50:53 -0700185 return NULL;
186}
187
188/**
189 * blkg_lookup - lookup blkg for the specified blkcg - q pair
190 * @blkcg: blkcg of interest
191 * @q: request_queue of interest
192 *
193 * Lookup blkg for the @blkcg - @q pair. This function should be called
194 * under RCU read lock and is guaranteed to return %NULL if @q is bypassing
195 * - see blk_queue_bypass_start() for details.
196 */
Tejun Heo3c798392012-04-16 13:57:25 -0700197struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q)
Tejun Heo80fd9972012-04-13 14:50:53 -0700198{
199 WARN_ON_ONCE(!rcu_read_lock_held());
200
201 if (unlikely(blk_queue_bypass(q)))
202 return NULL;
Tejun Heo86cde6b2013-01-09 08:05:10 -0800203 return __blkg_lookup(blkcg, q, false);
Tejun Heo80fd9972012-04-13 14:50:53 -0700204}
205EXPORT_SYMBOL_GPL(blkg_lookup);
206
Tejun Heo15974992012-06-04 20:40:52 -0700207/*
208 * If @new_blkg is %NULL, this function tries to allocate a new one as
209 * necessary using %GFP_ATOMIC. @new_blkg is always consumed on return.
210 */
Tejun Heo86cde6b2013-01-09 08:05:10 -0800211static struct blkcg_gq *blkg_create(struct blkcg *blkcg,
212 struct request_queue *q,
213 struct blkcg_gq *new_blkg)
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400214{
Tejun Heo3c798392012-04-16 13:57:25 -0700215 struct blkcg_gq *blkg;
Tejun Heof427d902013-01-09 08:05:12 -0800216 int i, ret;
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400217
Tejun Heocd1604f2012-03-05 13:15:06 -0800218 WARN_ON_ONCE(!rcu_read_lock_held());
219 lockdep_assert_held(q->queue_lock);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500220
Tejun Heo7ee9c562012-03-05 13:15:11 -0800221 /* blkg holds a reference to blkcg */
Tejun Heo15974992012-06-04 20:40:52 -0700222 if (!css_tryget(&blkcg->css)) {
Tejun Heo93e6d5d2013-01-09 08:05:10 -0800223 ret = -EINVAL;
224 goto err_free_blkg;
Tejun Heo15974992012-06-04 20:40:52 -0700225 }
Tejun Heocd1604f2012-03-05 13:15:06 -0800226
Tejun Heo496fb782012-04-19 16:29:23 -0700227 /* allocate */
Tejun Heo15974992012-06-04 20:40:52 -0700228 if (!new_blkg) {
229 new_blkg = blkg_alloc(blkcg, q, GFP_ATOMIC);
230 if (unlikely(!new_blkg)) {
Tejun Heo93e6d5d2013-01-09 08:05:10 -0800231 ret = -ENOMEM;
232 goto err_put_css;
Tejun Heo15974992012-06-04 20:40:52 -0700233 }
234 }
235 blkg = new_blkg;
Tejun Heocd1604f2012-03-05 13:15:06 -0800236
Tejun Heo3c547862013-01-09 08:05:10 -0800237 /* link parent and insert */
238 if (blkcg_parent(blkcg)) {
239 blkg->parent = __blkg_lookup(blkcg_parent(blkcg), q, false);
240 if (WARN_ON_ONCE(!blkg->parent)) {
241 blkg = ERR_PTR(-EINVAL);
242 goto err_put_css;
243 }
244 blkg_get(blkg->parent);
245 }
246
Tejun Heoa6371202012-04-19 16:29:24 -0700247 spin_lock(&blkcg->lock);
248 ret = radix_tree_insert(&blkcg->blkg_tree, q->id, blkg);
249 if (likely(!ret)) {
250 hlist_add_head_rcu(&blkg->blkcg_node, &blkcg->blkg_list);
251 list_add(&blkg->q_node, &q->blkg_list);
Tejun Heof427d902013-01-09 08:05:12 -0800252
253 for (i = 0; i < BLKCG_MAX_POLS; i++) {
254 struct blkcg_policy *pol = blkcg_policy[i];
255
256 if (blkg->pd[i] && pol->pd_online_fn)
257 pol->pd_online_fn(blkg);
258 }
Tejun Heoa6371202012-04-19 16:29:24 -0700259 }
Tejun Heof427d902013-01-09 08:05:12 -0800260 blkg->online = true;
Tejun Heoa6371202012-04-19 16:29:24 -0700261 spin_unlock(&blkcg->lock);
262
Tejun Heoa6371202012-04-19 16:29:24 -0700263 if (!ret)
264 return blkg;
Tejun Heo15974992012-06-04 20:40:52 -0700265
Tejun Heo3c547862013-01-09 08:05:10 -0800266 /* @blkg failed fully initialized, use the usual release path */
267 blkg_put(blkg);
268 return ERR_PTR(ret);
269
Tejun Heo93e6d5d2013-01-09 08:05:10 -0800270err_put_css:
Tejun Heo496fb782012-04-19 16:29:23 -0700271 css_put(&blkcg->css);
Tejun Heo93e6d5d2013-01-09 08:05:10 -0800272err_free_blkg:
Tejun Heo15974992012-06-04 20:40:52 -0700273 blkg_free(new_blkg);
Tejun Heo93e6d5d2013-01-09 08:05:10 -0800274 return ERR_PTR(ret);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500275}
Tejun Heo3c96cb32012-04-13 13:11:34 -0700276
Tejun Heo86cde6b2013-01-09 08:05:10 -0800277/**
278 * blkg_lookup_create - lookup blkg, try to create one if not there
279 * @blkcg: blkcg of interest
280 * @q: request_queue of interest
281 *
282 * Lookup blkg for the @blkcg - @q pair. If it doesn't exist, try to
Tejun Heo3c547862013-01-09 08:05:10 -0800283 * create one. blkg creation is performed recursively from blkcg_root such
284 * that all non-root blkg's have access to the parent blkg. This function
285 * should be called under RCU read lock and @q->queue_lock.
Tejun Heo86cde6b2013-01-09 08:05:10 -0800286 *
287 * Returns pointer to the looked up or created blkg on success, ERR_PTR()
288 * value on error. If @q is dead, returns ERR_PTR(-EINVAL). If @q is not
289 * dead and bypassing, returns ERR_PTR(-EBUSY).
290 */
Tejun Heo3c798392012-04-16 13:57:25 -0700291struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
292 struct request_queue *q)
Tejun Heo3c96cb32012-04-13 13:11:34 -0700293{
Tejun Heo86cde6b2013-01-09 08:05:10 -0800294 struct blkcg_gq *blkg;
295
296 WARN_ON_ONCE(!rcu_read_lock_held());
297 lockdep_assert_held(q->queue_lock);
298
Tejun Heo3c96cb32012-04-13 13:11:34 -0700299 /*
300 * This could be the first entry point of blkcg implementation and
301 * we shouldn't allow anything to go through for a bypassing queue.
302 */
303 if (unlikely(blk_queue_bypass(q)))
Bart Van Assche3f3299d2012-11-28 13:42:38 +0100304 return ERR_PTR(blk_queue_dying(q) ? -EINVAL : -EBUSY);
Tejun Heo86cde6b2013-01-09 08:05:10 -0800305
306 blkg = __blkg_lookup(blkcg, q, true);
307 if (blkg)
308 return blkg;
309
Tejun Heo3c547862013-01-09 08:05:10 -0800310 /*
311 * Create blkgs walking down from blkcg_root to @blkcg, so that all
312 * non-root blkgs have access to their parents.
313 */
314 while (true) {
315 struct blkcg *pos = blkcg;
316 struct blkcg *parent = blkcg_parent(blkcg);
317
318 while (parent && !__blkg_lookup(parent, q, false)) {
319 pos = parent;
320 parent = blkcg_parent(parent);
321 }
322
323 blkg = blkg_create(pos, q, NULL);
324 if (pos == blkcg || IS_ERR(blkg))
325 return blkg;
326 }
Tejun Heo3c96cb32012-04-13 13:11:34 -0700327}
Tejun Heocd1604f2012-03-05 13:15:06 -0800328EXPORT_SYMBOL_GPL(blkg_lookup_create);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500329
Tejun Heo3c798392012-04-16 13:57:25 -0700330static void blkg_destroy(struct blkcg_gq *blkg)
Tejun Heo72e06c22012-03-05 13:15:00 -0800331{
Tejun Heo3c798392012-04-16 13:57:25 -0700332 struct blkcg *blkcg = blkg->blkcg;
Tejun Heof427d902013-01-09 08:05:12 -0800333 int i;
Tejun Heo03aa2642012-03-05 13:15:19 -0800334
Tejun Heo27e1f9d2012-06-05 13:36:44 +0200335 lockdep_assert_held(blkg->q->queue_lock);
Tejun Heo9f13ef62012-03-05 13:15:21 -0800336 lockdep_assert_held(&blkcg->lock);
Tejun Heo03aa2642012-03-05 13:15:19 -0800337
338 /* Something wrong if we are trying to remove same group twice */
Tejun Heoe8989fa2012-03-05 13:15:20 -0800339 WARN_ON_ONCE(list_empty(&blkg->q_node));
Tejun Heo9f13ef62012-03-05 13:15:21 -0800340 WARN_ON_ONCE(hlist_unhashed(&blkg->blkcg_node));
Tejun Heoa6371202012-04-19 16:29:24 -0700341
Tejun Heof427d902013-01-09 08:05:12 -0800342 for (i = 0; i < BLKCG_MAX_POLS; i++) {
343 struct blkcg_policy *pol = blkcg_policy[i];
344
345 if (blkg->pd[i] && pol->pd_offline_fn)
346 pol->pd_offline_fn(blkg);
347 }
348 blkg->online = false;
349
Tejun Heoa6371202012-04-19 16:29:24 -0700350 radix_tree_delete(&blkcg->blkg_tree, blkg->q->id);
Tejun Heoe8989fa2012-03-05 13:15:20 -0800351 list_del_init(&blkg->q_node);
Tejun Heo9f13ef62012-03-05 13:15:21 -0800352 hlist_del_init_rcu(&blkg->blkcg_node);
Tejun Heo03aa2642012-03-05 13:15:19 -0800353
Tejun Heo03aa2642012-03-05 13:15:19 -0800354 /*
Tejun Heoa6371202012-04-19 16:29:24 -0700355 * Both setting lookup hint to and clearing it from @blkg are done
356 * under queue_lock. If it's not pointing to @blkg now, it never
357 * will. Hint assignment itself can race safely.
358 */
359 if (rcu_dereference_raw(blkcg->blkg_hint) == blkg)
360 rcu_assign_pointer(blkcg->blkg_hint, NULL);
361
362 /*
Tejun Heo03aa2642012-03-05 13:15:19 -0800363 * Put the reference taken at the time of creation so that when all
364 * queues are gone, group can be destroyed.
365 */
366 blkg_put(blkg);
367}
368
Tejun Heo9f13ef62012-03-05 13:15:21 -0800369/**
370 * blkg_destroy_all - destroy all blkgs associated with a request_queue
371 * @q: request_queue of interest
Tejun Heo9f13ef62012-03-05 13:15:21 -0800372 *
Tejun Heo3c96cb32012-04-13 13:11:34 -0700373 * Destroy all blkgs associated with @q.
Tejun Heo9f13ef62012-03-05 13:15:21 -0800374 */
Tejun Heo3c96cb32012-04-13 13:11:34 -0700375static void blkg_destroy_all(struct request_queue *q)
Tejun Heo03aa2642012-03-05 13:15:19 -0800376{
Tejun Heo3c798392012-04-16 13:57:25 -0700377 struct blkcg_gq *blkg, *n;
Tejun Heo72e06c22012-03-05 13:15:00 -0800378
Tejun Heo6d18b002012-04-13 13:11:35 -0700379 lockdep_assert_held(q->queue_lock);
Tejun Heo72e06c22012-03-05 13:15:00 -0800380
Tejun Heo9f13ef62012-03-05 13:15:21 -0800381 list_for_each_entry_safe(blkg, n, &q->blkg_list, q_node) {
Tejun Heo3c798392012-04-16 13:57:25 -0700382 struct blkcg *blkcg = blkg->blkcg;
Tejun Heo72e06c22012-03-05 13:15:00 -0800383
Tejun Heo9f13ef62012-03-05 13:15:21 -0800384 spin_lock(&blkcg->lock);
385 blkg_destroy(blkg);
386 spin_unlock(&blkcg->lock);
Tejun Heo72e06c22012-03-05 13:15:00 -0800387 }
Jun'ichi Nomura65635cb2012-10-17 17:45:36 +0900388
389 /*
390 * root blkg is destroyed. Just clear the pointer since
391 * root_rl does not take reference on root blkg.
392 */
393 q->root_blkg = NULL;
394 q->root_rl.blkg = NULL;
Tejun Heo72e06c22012-03-05 13:15:00 -0800395}
396
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800397static void blkg_rcu_free(struct rcu_head *rcu_head)
398{
Tejun Heo3c798392012-04-16 13:57:25 -0700399 blkg_free(container_of(rcu_head, struct blkcg_gq, rcu_head));
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800400}
401
Tejun Heo3c798392012-04-16 13:57:25 -0700402void __blkg_release(struct blkcg_gq *blkg)
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800403{
Tejun Heo3c547862013-01-09 08:05:10 -0800404 /* release the blkcg and parent blkg refs this blkg has been holding */
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800405 css_put(&blkg->blkcg->css);
Tejun Heo3c547862013-01-09 08:05:10 -0800406 if (blkg->parent)
407 blkg_put(blkg->parent);
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800408
409 /*
410 * A group is freed in rcu manner. But having an rcu lock does not
411 * mean that one can access all the fields of blkg and assume these
412 * are valid. For example, don't try to follow throtl_data and
413 * request queue links.
414 *
415 * Having a reference to blkg under an rcu allows acess to only
416 * values local to groups like group stats and group rate limits
417 */
418 call_rcu(&blkg->rcu_head, blkg_rcu_free);
419}
420EXPORT_SYMBOL_GPL(__blkg_release);
421
Tejun Heoa0516612012-06-26 15:05:44 -0700422/*
423 * The next function used by blk_queue_for_each_rl(). It's a bit tricky
424 * because the root blkg uses @q->root_rl instead of its own rl.
425 */
426struct request_list *__blk_queue_next_rl(struct request_list *rl,
427 struct request_queue *q)
428{
429 struct list_head *ent;
430 struct blkcg_gq *blkg;
431
432 /*
433 * Determine the current blkg list_head. The first entry is
434 * root_rl which is off @q->blkg_list and mapped to the head.
435 */
436 if (rl == &q->root_rl) {
437 ent = &q->blkg_list;
Jun'ichi Nomura65c77fd2012-10-22 10:15:37 +0900438 /* There are no more block groups, hence no request lists */
439 if (list_empty(ent))
440 return NULL;
Tejun Heoa0516612012-06-26 15:05:44 -0700441 } else {
442 blkg = container_of(rl, struct blkcg_gq, rl);
443 ent = &blkg->q_node;
444 }
445
446 /* walk to the next list_head, skip root blkcg */
447 ent = ent->next;
448 if (ent == &q->root_blkg->q_node)
449 ent = ent->next;
450 if (ent == &q->blkg_list)
451 return NULL;
452
453 blkg = container_of(ent, struct blkcg_gq, q_node);
454 return &blkg->rl;
455}
456
Tejun Heo3c798392012-04-16 13:57:25 -0700457static int blkcg_reset_stats(struct cgroup *cgroup, struct cftype *cftype,
458 u64 val)
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700459{
Tejun Heo3c798392012-04-16 13:57:25 -0700460 struct blkcg *blkcg = cgroup_to_blkcg(cgroup);
461 struct blkcg_gq *blkg;
Tejun Heobc0d6502012-04-13 13:11:26 -0700462 int i;
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700463
Tejun Heobc0d6502012-04-13 13:11:26 -0700464 mutex_lock(&blkcg_pol_mutex);
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700465 spin_lock_irq(&blkcg->lock);
Tejun Heo997a0262012-03-08 10:53:58 -0800466
467 /*
468 * Note that stat reset is racy - it doesn't synchronize against
469 * stat updates. This is a debug feature which shouldn't exist
470 * anyway. If you get hit by a race, retry.
471 */
Sasha Levinb67bfe02013-02-27 17:06:00 -0800472 hlist_for_each_entry(blkg, &blkcg->blkg_list, blkcg_node) {
Tejun Heo8bd435b2012-04-13 13:11:28 -0700473 for (i = 0; i < BLKCG_MAX_POLS; i++) {
Tejun Heo3c798392012-04-16 13:57:25 -0700474 struct blkcg_policy *pol = blkcg_policy[i];
Tejun Heo549d3aa2012-03-05 13:15:16 -0800475
Tejun Heoa2b16932012-04-13 13:11:33 -0700476 if (blkcg_policy_enabled(blkg->q, pol) &&
Tejun Heof9fcc2d2012-04-16 13:57:27 -0700477 pol->pd_reset_stats_fn)
478 pol->pd_reset_stats_fn(blkg);
Tejun Heobc0d6502012-04-13 13:11:26 -0700479 }
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700480 }
Vivek Goyalf0bdc8c2011-05-19 15:38:30 -0400481
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700482 spin_unlock_irq(&blkcg->lock);
Tejun Heobc0d6502012-04-13 13:11:26 -0700483 mutex_unlock(&blkcg_pol_mutex);
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700484 return 0;
485}
486
Tejun Heo3c798392012-04-16 13:57:25 -0700487static const char *blkg_dev_name(struct blkcg_gq *blkg)
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700488{
Tejun Heod3d32e62012-04-01 14:38:42 -0700489 /* some drivers (floppy) instantiate a queue w/o disk registered */
490 if (blkg->q->backing_dev_info.dev)
491 return dev_name(blkg->q->backing_dev_info.dev);
492 return NULL;
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700493}
494
Tejun Heod3d32e62012-04-01 14:38:42 -0700495/**
496 * blkcg_print_blkgs - helper for printing per-blkg data
497 * @sf: seq_file to print to
498 * @blkcg: blkcg of interest
499 * @prfill: fill function to print out a blkg
500 * @pol: policy in question
501 * @data: data to be passed to @prfill
502 * @show_total: to print out sum of prfill return values or not
503 *
504 * This function invokes @prfill on each blkg of @blkcg if pd for the
505 * policy specified by @pol exists. @prfill is invoked with @sf, the
Tejun Heo810ecfa2013-01-09 08:05:13 -0800506 * policy data and @data and the matching queue lock held. If @show_total
507 * is %true, the sum of the return values from @prfill is printed with
508 * "Total" label at the end.
Tejun Heod3d32e62012-04-01 14:38:42 -0700509 *
510 * This is to be used to construct print functions for
511 * cftype->read_seq_string method.
512 */
Tejun Heo3c798392012-04-16 13:57:25 -0700513void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
Tejun Heof95a04a2012-04-16 13:57:26 -0700514 u64 (*prfill)(struct seq_file *,
515 struct blkg_policy_data *, int),
Tejun Heo3c798392012-04-16 13:57:25 -0700516 const struct blkcg_policy *pol, int data,
Tejun Heoec399342012-04-13 13:11:27 -0700517 bool show_total)
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400518{
Tejun Heo3c798392012-04-16 13:57:25 -0700519 struct blkcg_gq *blkg;
Tejun Heod3d32e62012-04-01 14:38:42 -0700520 u64 total = 0;
521
Tejun Heo810ecfa2013-01-09 08:05:13 -0800522 rcu_read_lock();
Linus Torvaldsee89f812013-02-28 12:52:24 -0800523 hlist_for_each_entry_rcu(blkg, &blkcg->blkg_list, blkcg_node) {
Tejun Heo810ecfa2013-01-09 08:05:13 -0800524 spin_lock_irq(blkg->q->queue_lock);
Tejun Heoa2b16932012-04-13 13:11:33 -0700525 if (blkcg_policy_enabled(blkg->q, pol))
Tejun Heof95a04a2012-04-16 13:57:26 -0700526 total += prfill(sf, blkg->pd[pol->plid], data);
Tejun Heo810ecfa2013-01-09 08:05:13 -0800527 spin_unlock_irq(blkg->q->queue_lock);
528 }
529 rcu_read_unlock();
Tejun Heod3d32e62012-04-01 14:38:42 -0700530
531 if (show_total)
532 seq_printf(sf, "Total %llu\n", (unsigned long long)total);
533}
Tejun Heo829fdb52012-04-01 14:38:43 -0700534EXPORT_SYMBOL_GPL(blkcg_print_blkgs);
Tejun Heod3d32e62012-04-01 14:38:42 -0700535
536/**
537 * __blkg_prfill_u64 - prfill helper for a single u64 value
538 * @sf: seq_file to print to
Tejun Heof95a04a2012-04-16 13:57:26 -0700539 * @pd: policy private data of interest
Tejun Heod3d32e62012-04-01 14:38:42 -0700540 * @v: value to print
541 *
Tejun Heof95a04a2012-04-16 13:57:26 -0700542 * Print @v to @sf for the device assocaited with @pd.
Tejun Heod3d32e62012-04-01 14:38:42 -0700543 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700544u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v)
Tejun Heod3d32e62012-04-01 14:38:42 -0700545{
Tejun Heof95a04a2012-04-16 13:57:26 -0700546 const char *dname = blkg_dev_name(pd->blkg);
Tejun Heod3d32e62012-04-01 14:38:42 -0700547
548 if (!dname)
549 return 0;
550
551 seq_printf(sf, "%s %llu\n", dname, (unsigned long long)v);
552 return v;
553}
Tejun Heo829fdb52012-04-01 14:38:43 -0700554EXPORT_SYMBOL_GPL(__blkg_prfill_u64);
Tejun Heod3d32e62012-04-01 14:38:42 -0700555
556/**
557 * __blkg_prfill_rwstat - prfill helper for a blkg_rwstat
558 * @sf: seq_file to print to
Tejun Heof95a04a2012-04-16 13:57:26 -0700559 * @pd: policy private data of interest
Tejun Heod3d32e62012-04-01 14:38:42 -0700560 * @rwstat: rwstat to print
561 *
Tejun Heof95a04a2012-04-16 13:57:26 -0700562 * Print @rwstat to @sf for the device assocaited with @pd.
Tejun Heod3d32e62012-04-01 14:38:42 -0700563 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700564u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
Tejun Heo829fdb52012-04-01 14:38:43 -0700565 const struct blkg_rwstat *rwstat)
Tejun Heod3d32e62012-04-01 14:38:42 -0700566{
567 static const char *rwstr[] = {
568 [BLKG_RWSTAT_READ] = "Read",
569 [BLKG_RWSTAT_WRITE] = "Write",
570 [BLKG_RWSTAT_SYNC] = "Sync",
571 [BLKG_RWSTAT_ASYNC] = "Async",
572 };
Tejun Heof95a04a2012-04-16 13:57:26 -0700573 const char *dname = blkg_dev_name(pd->blkg);
Tejun Heod3d32e62012-04-01 14:38:42 -0700574 u64 v;
575 int i;
576
577 if (!dname)
578 return 0;
579
580 for (i = 0; i < BLKG_RWSTAT_NR; i++)
581 seq_printf(sf, "%s %s %llu\n", dname, rwstr[i],
582 (unsigned long long)rwstat->cnt[i]);
583
584 v = rwstat->cnt[BLKG_RWSTAT_READ] + rwstat->cnt[BLKG_RWSTAT_WRITE];
585 seq_printf(sf, "%s Total %llu\n", dname, (unsigned long long)v);
586 return v;
587}
Tejun Heob50da392013-01-09 08:05:12 -0800588EXPORT_SYMBOL_GPL(__blkg_prfill_rwstat);
Tejun Heod3d32e62012-04-01 14:38:42 -0700589
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700590/**
591 * blkg_prfill_stat - prfill callback for blkg_stat
592 * @sf: seq_file to print to
Tejun Heof95a04a2012-04-16 13:57:26 -0700593 * @pd: policy private data of interest
594 * @off: offset to the blkg_stat in @pd
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700595 *
596 * prfill callback for printing a blkg_stat.
597 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700598u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off)
Tejun Heod3d32e62012-04-01 14:38:42 -0700599{
Tejun Heof95a04a2012-04-16 13:57:26 -0700600 return __blkg_prfill_u64(sf, pd, blkg_stat_read((void *)pd + off));
Tejun Heod3d32e62012-04-01 14:38:42 -0700601}
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700602EXPORT_SYMBOL_GPL(blkg_prfill_stat);
Tejun Heod3d32e62012-04-01 14:38:42 -0700603
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700604/**
605 * blkg_prfill_rwstat - prfill callback for blkg_rwstat
606 * @sf: seq_file to print to
Tejun Heof95a04a2012-04-16 13:57:26 -0700607 * @pd: policy private data of interest
608 * @off: offset to the blkg_rwstat in @pd
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700609 *
610 * prfill callback for printing a blkg_rwstat.
611 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700612u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
613 int off)
Tejun Heod3d32e62012-04-01 14:38:42 -0700614{
Tejun Heof95a04a2012-04-16 13:57:26 -0700615 struct blkg_rwstat rwstat = blkg_rwstat_read((void *)pd + off);
Tejun Heod3d32e62012-04-01 14:38:42 -0700616
Tejun Heof95a04a2012-04-16 13:57:26 -0700617 return __blkg_prfill_rwstat(sf, pd, &rwstat);
Tejun Heod3d32e62012-04-01 14:38:42 -0700618}
Tejun Heo5bc4afb12012-04-01 14:38:45 -0700619EXPORT_SYMBOL_GPL(blkg_prfill_rwstat);
Tejun Heod3d32e62012-04-01 14:38:42 -0700620
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700621/**
Tejun Heo16b3de62013-01-09 08:05:12 -0800622 * blkg_stat_recursive_sum - collect hierarchical blkg_stat
623 * @pd: policy private data of interest
624 * @off: offset to the blkg_stat in @pd
625 *
626 * Collect the blkg_stat specified by @off from @pd and all its online
627 * descendants and return the sum. The caller must be holding the queue
628 * lock for online tests.
629 */
630u64 blkg_stat_recursive_sum(struct blkg_policy_data *pd, int off)
631{
632 struct blkcg_policy *pol = blkcg_policy[pd->plid];
633 struct blkcg_gq *pos_blkg;
634 struct cgroup *pos_cgrp;
635 u64 sum;
636
637 lockdep_assert_held(pd->blkg->q->queue_lock);
638
639 sum = blkg_stat_read((void *)pd + off);
640
641 rcu_read_lock();
642 blkg_for_each_descendant_pre(pos_blkg, pos_cgrp, pd_to_blkg(pd)) {
643 struct blkg_policy_data *pos_pd = blkg_to_pd(pos_blkg, pol);
644 struct blkg_stat *stat = (void *)pos_pd + off;
645
646 if (pos_blkg->online)
647 sum += blkg_stat_read(stat);
648 }
649 rcu_read_unlock();
650
651 return sum;
652}
653EXPORT_SYMBOL_GPL(blkg_stat_recursive_sum);
654
655/**
656 * blkg_rwstat_recursive_sum - collect hierarchical blkg_rwstat
657 * @pd: policy private data of interest
658 * @off: offset to the blkg_stat in @pd
659 *
660 * Collect the blkg_rwstat specified by @off from @pd and all its online
661 * descendants and return the sum. The caller must be holding the queue
662 * lock for online tests.
663 */
664struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkg_policy_data *pd,
665 int off)
666{
667 struct blkcg_policy *pol = blkcg_policy[pd->plid];
668 struct blkcg_gq *pos_blkg;
669 struct cgroup *pos_cgrp;
670 struct blkg_rwstat sum;
671 int i;
672
673 lockdep_assert_held(pd->blkg->q->queue_lock);
674
675 sum = blkg_rwstat_read((void *)pd + off);
676
677 rcu_read_lock();
678 blkg_for_each_descendant_pre(pos_blkg, pos_cgrp, pd_to_blkg(pd)) {
679 struct blkg_policy_data *pos_pd = blkg_to_pd(pos_blkg, pol);
680 struct blkg_rwstat *rwstat = (void *)pos_pd + off;
681 struct blkg_rwstat tmp;
682
683 if (!pos_blkg->online)
684 continue;
685
686 tmp = blkg_rwstat_read(rwstat);
687
688 for (i = 0; i < BLKG_RWSTAT_NR; i++)
689 sum.cnt[i] += tmp.cnt[i];
690 }
691 rcu_read_unlock();
692
693 return sum;
694}
695EXPORT_SYMBOL_GPL(blkg_rwstat_recursive_sum);
696
697/**
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700698 * blkg_conf_prep - parse and prepare for per-blkg config update
699 * @blkcg: target block cgroup
Tejun Heoda8b0662012-04-13 13:11:29 -0700700 * @pol: target policy
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700701 * @input: input string
702 * @ctx: blkg_conf_ctx to be filled
703 *
704 * Parse per-blkg config update from @input and initialize @ctx with the
705 * result. @ctx->blkg points to the blkg to be updated and @ctx->v the new
Tejun Heoda8b0662012-04-13 13:11:29 -0700706 * value. This function returns with RCU read lock and queue lock held and
707 * must be paired with blkg_conf_finish().
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700708 */
Tejun Heo3c798392012-04-16 13:57:25 -0700709int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
710 const char *input, struct blkg_conf_ctx *ctx)
Tejun Heoda8b0662012-04-13 13:11:29 -0700711 __acquires(rcu) __acquires(disk->queue->queue_lock)
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800712{
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700713 struct gendisk *disk;
Tejun Heo3c798392012-04-16 13:57:25 -0700714 struct blkcg_gq *blkg;
Tejun Heo726fa692012-04-01 14:38:43 -0700715 unsigned int major, minor;
716 unsigned long long v;
717 int part, ret;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800718
Tejun Heo726fa692012-04-01 14:38:43 -0700719 if (sscanf(input, "%u:%u %llu", &major, &minor, &v) != 3)
720 return -EINVAL;
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700721
Tejun Heo726fa692012-04-01 14:38:43 -0700722 disk = get_gendisk(MKDEV(major, minor), &part);
Tejun Heo4bfd4822012-03-05 13:15:08 -0800723 if (!disk || part)
Tejun Heo726fa692012-04-01 14:38:43 -0700724 return -EINVAL;
Tejun Heoe56da7e2012-03-05 13:15:07 -0800725
726 rcu_read_lock();
Tejun Heo4bfd4822012-03-05 13:15:08 -0800727 spin_lock_irq(disk->queue->queue_lock);
Tejun Heoda8b0662012-04-13 13:11:29 -0700728
Tejun Heoa2b16932012-04-13 13:11:33 -0700729 if (blkcg_policy_enabled(disk->queue, pol))
Tejun Heo3c96cb32012-04-13 13:11:34 -0700730 blkg = blkg_lookup_create(blkcg, disk->queue);
Tejun Heoa2b16932012-04-13 13:11:33 -0700731 else
732 blkg = ERR_PTR(-EINVAL);
Tejun Heoe56da7e2012-03-05 13:15:07 -0800733
Tejun Heo4bfd4822012-03-05 13:15:08 -0800734 if (IS_ERR(blkg)) {
735 ret = PTR_ERR(blkg);
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700736 rcu_read_unlock();
Tejun Heoda8b0662012-04-13 13:11:29 -0700737 spin_unlock_irq(disk->queue->queue_lock);
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700738 put_disk(disk);
739 /*
740 * If queue was bypassing, we should retry. Do so after a
741 * short msleep(). It isn't strictly necessary but queue
742 * can be bypassing for some time and it's always nice to
743 * avoid busy looping.
744 */
745 if (ret == -EBUSY) {
746 msleep(10);
747 ret = restart_syscall();
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400748 }
Tejun Heo726fa692012-04-01 14:38:43 -0700749 return ret;
Vivek Goyal062a6442010-09-15 17:06:33 -0400750 }
Tejun Heoe56da7e2012-03-05 13:15:07 -0800751
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700752 ctx->disk = disk;
753 ctx->blkg = blkg;
Tejun Heo726fa692012-04-01 14:38:43 -0700754 ctx->v = v;
755 return 0;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800756}
Tejun Heo829fdb52012-04-01 14:38:43 -0700757EXPORT_SYMBOL_GPL(blkg_conf_prep);
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800758
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700759/**
760 * blkg_conf_finish - finish up per-blkg config update
761 * @ctx: blkg_conf_ctx intiailized by blkg_conf_prep()
762 *
763 * Finish up after per-blkg config update. This function must be paired
764 * with blkg_conf_prep().
765 */
Tejun Heo829fdb52012-04-01 14:38:43 -0700766void blkg_conf_finish(struct blkg_conf_ctx *ctx)
Tejun Heoda8b0662012-04-13 13:11:29 -0700767 __releases(ctx->disk->queue->queue_lock) __releases(rcu)
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800768{
Tejun Heoda8b0662012-04-13 13:11:29 -0700769 spin_unlock_irq(ctx->disk->queue->queue_lock);
Tejun Heo3a8b31d2012-04-01 14:38:43 -0700770 rcu_read_unlock();
771 put_disk(ctx->disk);
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800772}
Tejun Heo829fdb52012-04-01 14:38:43 -0700773EXPORT_SYMBOL_GPL(blkg_conf_finish);
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800774
Tejun Heo3c798392012-04-16 13:57:25 -0700775struct cftype blkcg_files[] = {
Vivek Goyal31e4c282009-12-03 12:59:42 -0500776 {
Divyesh Shah84c124d2010-04-09 08:31:19 +0200777 .name = "reset_stats",
Tejun Heo3c798392012-04-16 13:57:25 -0700778 .write_u64 = blkcg_reset_stats,
Vivek Goyal22084192009-12-03 12:59:49 -0500779 },
Tejun Heo4baf6e32012-04-01 12:09:55 -0700780 { } /* terminate */
Vivek Goyal31e4c282009-12-03 12:59:42 -0500781};
782
Tejun Heo9f13ef62012-03-05 13:15:21 -0800783/**
Tejun Heo92fb9742012-11-19 08:13:38 -0800784 * blkcg_css_offline - cgroup css_offline callback
Tejun Heo9f13ef62012-03-05 13:15:21 -0800785 * @cgroup: cgroup of interest
786 *
787 * This function is called when @cgroup is about to go away and responsible
788 * for shooting down all blkgs associated with @cgroup. blkgs should be
789 * removed while holding both q and blkcg locks. As blkcg lock is nested
790 * inside q lock, this function performs reverse double lock dancing.
791 *
792 * This is the blkcg counterpart of ioc_release_fn().
793 */
Tejun Heo92fb9742012-11-19 08:13:38 -0800794static void blkcg_css_offline(struct cgroup *cgroup)
Vivek Goyal31e4c282009-12-03 12:59:42 -0500795{
Tejun Heo3c798392012-04-16 13:57:25 -0700796 struct blkcg *blkcg = cgroup_to_blkcg(cgroup);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500797
Tejun Heo9f13ef62012-03-05 13:15:21 -0800798 spin_lock_irq(&blkcg->lock);
Tejun Heo7ee9c562012-03-05 13:15:11 -0800799
Tejun Heo9f13ef62012-03-05 13:15:21 -0800800 while (!hlist_empty(&blkcg->blkg_list)) {
Tejun Heo3c798392012-04-16 13:57:25 -0700801 struct blkcg_gq *blkg = hlist_entry(blkcg->blkg_list.first,
802 struct blkcg_gq, blkcg_node);
Tejun Heoc875f4d2012-03-05 13:15:22 -0800803 struct request_queue *q = blkg->q;
Vivek Goyalb1c35762009-12-03 12:59:47 -0500804
Tejun Heo9f13ef62012-03-05 13:15:21 -0800805 if (spin_trylock(q->queue_lock)) {
806 blkg_destroy(blkg);
807 spin_unlock(q->queue_lock);
808 } else {
809 spin_unlock_irq(&blkcg->lock);
Tejun Heo9f13ef62012-03-05 13:15:21 -0800810 cpu_relax();
Dan Carpentera5567932012-03-29 20:57:08 +0200811 spin_lock_irq(&blkcg->lock);
Jens Axboe0f3942a2010-05-03 14:28:55 +0200812 }
Tejun Heo9f13ef62012-03-05 13:15:21 -0800813 }
Jens Axboe0f3942a2010-05-03 14:28:55 +0200814
Tejun Heo9f13ef62012-03-05 13:15:21 -0800815 spin_unlock_irq(&blkcg->lock);
Tejun Heo7ee9c562012-03-05 13:15:11 -0800816}
817
Tejun Heo92fb9742012-11-19 08:13:38 -0800818static void blkcg_css_free(struct cgroup *cgroup)
Tejun Heo7ee9c562012-03-05 13:15:11 -0800819{
Tejun Heo3c798392012-04-16 13:57:25 -0700820 struct blkcg *blkcg = cgroup_to_blkcg(cgroup);
Tejun Heo7ee9c562012-03-05 13:15:11 -0800821
Tejun Heo3c798392012-04-16 13:57:25 -0700822 if (blkcg != &blkcg_root)
Ben Blum67523c42010-03-10 15:22:11 -0800823 kfree(blkcg);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500824}
825
Tejun Heo92fb9742012-11-19 08:13:38 -0800826static struct cgroup_subsys_state *blkcg_css_alloc(struct cgroup *cgroup)
Vivek Goyal31e4c282009-12-03 12:59:42 -0500827{
Tejun Heo9a9e8a22012-03-19 15:10:56 -0700828 static atomic64_t id_seq = ATOMIC64_INIT(0);
Tejun Heo3c798392012-04-16 13:57:25 -0700829 struct blkcg *blkcg;
Li Zefan03415092010-05-07 08:57:00 +0200830 struct cgroup *parent = cgroup->parent;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500831
Li Zefan03415092010-05-07 08:57:00 +0200832 if (!parent) {
Tejun Heo3c798392012-04-16 13:57:25 -0700833 blkcg = &blkcg_root;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500834 goto done;
835 }
836
Vivek Goyal31e4c282009-12-03 12:59:42 -0500837 blkcg = kzalloc(sizeof(*blkcg), GFP_KERNEL);
838 if (!blkcg)
839 return ERR_PTR(-ENOMEM);
840
Tejun Heo3381cb82012-04-01 14:38:44 -0700841 blkcg->cfq_weight = CFQ_WEIGHT_DEFAULT;
Tejun Heoe71357e2013-01-09 08:05:10 -0800842 blkcg->cfq_leaf_weight = CFQ_WEIGHT_DEFAULT;
Tejun Heo9a9e8a22012-03-19 15:10:56 -0700843 blkcg->id = atomic64_inc_return(&id_seq); /* root is 0, start from 1 */
Vivek Goyal31e4c282009-12-03 12:59:42 -0500844done:
845 spin_lock_init(&blkcg->lock);
Tejun Heoa6371202012-04-19 16:29:24 -0700846 INIT_RADIX_TREE(&blkcg->blkg_tree, GFP_ATOMIC);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500847 INIT_HLIST_HEAD(&blkcg->blkg_list);
848
849 return &blkcg->css;
850}
851
Tejun Heo5efd6112012-03-05 13:15:12 -0800852/**
853 * blkcg_init_queue - initialize blkcg part of request queue
854 * @q: request_queue to initialize
855 *
856 * Called from blk_alloc_queue_node(). Responsible for initializing blkcg
857 * part of new request_queue @q.
858 *
859 * RETURNS:
860 * 0 on success, -errno on failure.
861 */
862int blkcg_init_queue(struct request_queue *q)
863{
864 might_sleep();
865
Tejun Heo3c96cb32012-04-13 13:11:34 -0700866 return blk_throtl_init(q);
Tejun Heo5efd6112012-03-05 13:15:12 -0800867}
868
869/**
870 * blkcg_drain_queue - drain blkcg part of request_queue
871 * @q: request_queue to drain
872 *
873 * Called from blk_drain_queue(). Responsible for draining blkcg part.
874 */
875void blkcg_drain_queue(struct request_queue *q)
876{
877 lockdep_assert_held(q->queue_lock);
878
879 blk_throtl_drain(q);
880}
881
882/**
883 * blkcg_exit_queue - exit and release blkcg part of request_queue
884 * @q: request_queue being released
885 *
886 * Called from blk_release_queue(). Responsible for exiting blkcg part.
887 */
888void blkcg_exit_queue(struct request_queue *q)
889{
Tejun Heo6d18b002012-04-13 13:11:35 -0700890 spin_lock_irq(q->queue_lock);
Tejun Heo3c96cb32012-04-13 13:11:34 -0700891 blkg_destroy_all(q);
Tejun Heo6d18b002012-04-13 13:11:35 -0700892 spin_unlock_irq(q->queue_lock);
893
Tejun Heo5efd6112012-03-05 13:15:12 -0800894 blk_throtl_exit(q);
895}
896
Vivek Goyal31e4c282009-12-03 12:59:42 -0500897/*
898 * We cannot support shared io contexts, as we have no mean to support
899 * two tasks with the same ioc in two different groups without major rework
900 * of the main cic data structures. For now we allow a task to change
901 * its cgroup only if it's the only owner of its ioc.
902 */
Tejun Heo3c798392012-04-16 13:57:25 -0700903static int blkcg_can_attach(struct cgroup *cgrp, struct cgroup_taskset *tset)
Vivek Goyal31e4c282009-12-03 12:59:42 -0500904{
Tejun Heobb9d97b2011-12-12 18:12:21 -0800905 struct task_struct *task;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500906 struct io_context *ioc;
907 int ret = 0;
908
909 /* task_lock() is needed to avoid races with exit_io_context() */
Tejun Heobb9d97b2011-12-12 18:12:21 -0800910 cgroup_taskset_for_each(task, cgrp, tset) {
911 task_lock(task);
912 ioc = task->io_context;
913 if (ioc && atomic_read(&ioc->nr_tasks) > 1)
914 ret = -EINVAL;
915 task_unlock(task);
916 if (ret)
917 break;
918 }
Vivek Goyal31e4c282009-12-03 12:59:42 -0500919 return ret;
920}
921
Tejun Heo676f7c82012-04-01 12:09:55 -0700922struct cgroup_subsys blkio_subsys = {
923 .name = "blkio",
Tejun Heo92fb9742012-11-19 08:13:38 -0800924 .css_alloc = blkcg_css_alloc,
925 .css_offline = blkcg_css_offline,
926 .css_free = blkcg_css_free,
Tejun Heo3c798392012-04-16 13:57:25 -0700927 .can_attach = blkcg_can_attach,
Tejun Heo676f7c82012-04-01 12:09:55 -0700928 .subsys_id = blkio_subsys_id,
Tejun Heo3c798392012-04-16 13:57:25 -0700929 .base_cftypes = blkcg_files,
Tejun Heo676f7c82012-04-01 12:09:55 -0700930 .module = THIS_MODULE,
Tejun Heo8c7f6ed2012-09-13 12:20:58 -0700931
932 /*
933 * blkio subsystem is utterly broken in terms of hierarchy support.
934 * It treats all cgroups equally regardless of where they're
935 * located in the hierarchy - all cgroups are treated as if they're
936 * right below the root. Fix it and remove the following.
937 */
938 .broken_hierarchy = true,
Tejun Heo676f7c82012-04-01 12:09:55 -0700939};
940EXPORT_SYMBOL_GPL(blkio_subsys);
941
Tejun Heo8bd435b2012-04-13 13:11:28 -0700942/**
Tejun Heoa2b16932012-04-13 13:11:33 -0700943 * blkcg_activate_policy - activate a blkcg policy on a request_queue
944 * @q: request_queue of interest
945 * @pol: blkcg policy to activate
946 *
947 * Activate @pol on @q. Requires %GFP_KERNEL context. @q goes through
948 * bypass mode to populate its blkgs with policy_data for @pol.
949 *
950 * Activation happens with @q bypassed, so nobody would be accessing blkgs
951 * from IO path. Update of each blkg is protected by both queue and blkcg
952 * locks so that holding either lock and testing blkcg_policy_enabled() is
953 * always enough for dereferencing policy data.
954 *
955 * The caller is responsible for synchronizing [de]activations and policy
956 * [un]registerations. Returns 0 on success, -errno on failure.
957 */
958int blkcg_activate_policy(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -0700959 const struct blkcg_policy *pol)
Tejun Heoa2b16932012-04-13 13:11:33 -0700960{
961 LIST_HEAD(pds);
Tejun Heo86cde6b2013-01-09 08:05:10 -0800962 struct blkcg_gq *blkg, *new_blkg;
Tejun Heoa2b16932012-04-13 13:11:33 -0700963 struct blkg_policy_data *pd, *n;
964 int cnt = 0, ret;
Tejun Heo15974992012-06-04 20:40:52 -0700965 bool preloaded;
Tejun Heoa2b16932012-04-13 13:11:33 -0700966
967 if (blkcg_policy_enabled(q, pol))
968 return 0;
969
Tejun Heo15974992012-06-04 20:40:52 -0700970 /* preallocations for root blkg */
Tejun Heo86cde6b2013-01-09 08:05:10 -0800971 new_blkg = blkg_alloc(&blkcg_root, q, GFP_KERNEL);
972 if (!new_blkg)
Tejun Heo15974992012-06-04 20:40:52 -0700973 return -ENOMEM;
974
Tejun Heoa2b16932012-04-13 13:11:33 -0700975 blk_queue_bypass_start(q);
976
Jun'ichi Nomurae5072662013-04-09 15:01:21 +0200977 preloaded = !radix_tree_preload(GFP_KERNEL);
978
Tejun Heo86cde6b2013-01-09 08:05:10 -0800979 /*
980 * Make sure the root blkg exists and count the existing blkgs. As
981 * @q is bypassing at this point, blkg_lookup_create() can't be
982 * used. Open code it.
983 */
Tejun Heoa2b16932012-04-13 13:11:33 -0700984 spin_lock_irq(q->queue_lock);
985
986 rcu_read_lock();
Tejun Heo86cde6b2013-01-09 08:05:10 -0800987 blkg = __blkg_lookup(&blkcg_root, q, false);
988 if (blkg)
989 blkg_free(new_blkg);
990 else
991 blkg = blkg_create(&blkcg_root, q, new_blkg);
Tejun Heoa2b16932012-04-13 13:11:33 -0700992 rcu_read_unlock();
993
Tejun Heo15974992012-06-04 20:40:52 -0700994 if (preloaded)
995 radix_tree_preload_end();
996
Tejun Heoa2b16932012-04-13 13:11:33 -0700997 if (IS_ERR(blkg)) {
998 ret = PTR_ERR(blkg);
999 goto out_unlock;
1000 }
1001 q->root_blkg = blkg;
Tejun Heoa0516612012-06-26 15:05:44 -07001002 q->root_rl.blkg = blkg;
Tejun Heoa2b16932012-04-13 13:11:33 -07001003
1004 list_for_each_entry(blkg, &q->blkg_list, q_node)
1005 cnt++;
1006
1007 spin_unlock_irq(q->queue_lock);
1008
1009 /* allocate policy_data for all existing blkgs */
1010 while (cnt--) {
Tejun Heof95a04a2012-04-16 13:57:26 -07001011 pd = kzalloc_node(pol->pd_size, GFP_KERNEL, q->node);
Tejun Heoa2b16932012-04-13 13:11:33 -07001012 if (!pd) {
1013 ret = -ENOMEM;
1014 goto out_free;
1015 }
1016 list_add_tail(&pd->alloc_node, &pds);
1017 }
1018
1019 /*
1020 * Install the allocated pds. With @q bypassing, no new blkg
1021 * should have been created while the queue lock was dropped.
1022 */
1023 spin_lock_irq(q->queue_lock);
1024
1025 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1026 if (WARN_ON(list_empty(&pds))) {
1027 /* umm... this shouldn't happen, just abort */
1028 ret = -ENOMEM;
1029 goto out_unlock;
1030 }
1031 pd = list_first_entry(&pds, struct blkg_policy_data, alloc_node);
1032 list_del_init(&pd->alloc_node);
1033
1034 /* grab blkcg lock too while installing @pd on @blkg */
1035 spin_lock(&blkg->blkcg->lock);
1036
1037 blkg->pd[pol->plid] = pd;
1038 pd->blkg = blkg;
Tejun Heob276a872013-01-09 08:05:12 -08001039 pd->plid = pol->plid;
Tejun Heof9fcc2d2012-04-16 13:57:27 -07001040 pol->pd_init_fn(blkg);
Tejun Heoa2b16932012-04-13 13:11:33 -07001041
1042 spin_unlock(&blkg->blkcg->lock);
1043 }
1044
1045 __set_bit(pol->plid, q->blkcg_pols);
1046 ret = 0;
1047out_unlock:
1048 spin_unlock_irq(q->queue_lock);
1049out_free:
1050 blk_queue_bypass_end(q);
1051 list_for_each_entry_safe(pd, n, &pds, alloc_node)
1052 kfree(pd);
1053 return ret;
1054}
1055EXPORT_SYMBOL_GPL(blkcg_activate_policy);
1056
1057/**
1058 * blkcg_deactivate_policy - deactivate a blkcg policy on a request_queue
1059 * @q: request_queue of interest
1060 * @pol: blkcg policy to deactivate
1061 *
1062 * Deactivate @pol on @q. Follows the same synchronization rules as
1063 * blkcg_activate_policy().
1064 */
1065void blkcg_deactivate_policy(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -07001066 const struct blkcg_policy *pol)
Tejun Heoa2b16932012-04-13 13:11:33 -07001067{
Tejun Heo3c798392012-04-16 13:57:25 -07001068 struct blkcg_gq *blkg;
Tejun Heoa2b16932012-04-13 13:11:33 -07001069
1070 if (!blkcg_policy_enabled(q, pol))
1071 return;
1072
1073 blk_queue_bypass_start(q);
1074 spin_lock_irq(q->queue_lock);
1075
1076 __clear_bit(pol->plid, q->blkcg_pols);
1077
Tejun Heo6d18b002012-04-13 13:11:35 -07001078 /* if no policy is left, no need for blkgs - shoot them down */
1079 if (bitmap_empty(q->blkcg_pols, BLKCG_MAX_POLS))
1080 blkg_destroy_all(q);
1081
Tejun Heoa2b16932012-04-13 13:11:33 -07001082 list_for_each_entry(blkg, &q->blkg_list, q_node) {
1083 /* grab blkcg lock too while removing @pd from @blkg */
1084 spin_lock(&blkg->blkcg->lock);
1085
Tejun Heof427d902013-01-09 08:05:12 -08001086 if (pol->pd_offline_fn)
1087 pol->pd_offline_fn(blkg);
Tejun Heof9fcc2d2012-04-16 13:57:27 -07001088 if (pol->pd_exit_fn)
1089 pol->pd_exit_fn(blkg);
Tejun Heoa2b16932012-04-13 13:11:33 -07001090
1091 kfree(blkg->pd[pol->plid]);
1092 blkg->pd[pol->plid] = NULL;
1093
1094 spin_unlock(&blkg->blkcg->lock);
1095 }
1096
1097 spin_unlock_irq(q->queue_lock);
1098 blk_queue_bypass_end(q);
1099}
1100EXPORT_SYMBOL_GPL(blkcg_deactivate_policy);
1101
1102/**
Tejun Heo3c798392012-04-16 13:57:25 -07001103 * blkcg_policy_register - register a blkcg policy
1104 * @pol: blkcg policy to register
Tejun Heo8bd435b2012-04-13 13:11:28 -07001105 *
Tejun Heo3c798392012-04-16 13:57:25 -07001106 * Register @pol with blkcg core. Might sleep and @pol may be modified on
1107 * successful registration. Returns 0 on success and -errno on failure.
Tejun Heo8bd435b2012-04-13 13:11:28 -07001108 */
Tejun Heo3c798392012-04-16 13:57:25 -07001109int blkcg_policy_register(struct blkcg_policy *pol)
Vivek Goyal3e252062009-12-04 10:36:42 -05001110{
Tejun Heo8bd435b2012-04-13 13:11:28 -07001111 int i, ret;
Tejun Heoe8989fa2012-03-05 13:15:20 -08001112
Tejun Heof95a04a2012-04-16 13:57:26 -07001113 if (WARN_ON(pol->pd_size < sizeof(struct blkg_policy_data)))
1114 return -EINVAL;
1115
Tejun Heobc0d6502012-04-13 13:11:26 -07001116 mutex_lock(&blkcg_pol_mutex);
1117
Tejun Heo8bd435b2012-04-13 13:11:28 -07001118 /* find an empty slot */
1119 ret = -ENOSPC;
1120 for (i = 0; i < BLKCG_MAX_POLS; i++)
Tejun Heo3c798392012-04-16 13:57:25 -07001121 if (!blkcg_policy[i])
Tejun Heo8bd435b2012-04-13 13:11:28 -07001122 break;
1123 if (i >= BLKCG_MAX_POLS)
1124 goto out_unlock;
Tejun Heo035d10b2012-03-05 13:15:04 -08001125
Tejun Heo8bd435b2012-04-13 13:11:28 -07001126 /* register and update blkgs */
Tejun Heo3c798392012-04-16 13:57:25 -07001127 pol->plid = i;
1128 blkcg_policy[i] = pol;
Tejun Heo8bd435b2012-04-13 13:11:28 -07001129
Tejun Heo8bd435b2012-04-13 13:11:28 -07001130 /* everything is in place, add intf files for the new policy */
Tejun Heo3c798392012-04-16 13:57:25 -07001131 if (pol->cftypes)
1132 WARN_ON(cgroup_add_cftypes(&blkio_subsys, pol->cftypes));
Tejun Heo8bd435b2012-04-13 13:11:28 -07001133 ret = 0;
1134out_unlock:
Tejun Heobc0d6502012-04-13 13:11:26 -07001135 mutex_unlock(&blkcg_pol_mutex);
Tejun Heo8bd435b2012-04-13 13:11:28 -07001136 return ret;
Vivek Goyal3e252062009-12-04 10:36:42 -05001137}
Tejun Heo3c798392012-04-16 13:57:25 -07001138EXPORT_SYMBOL_GPL(blkcg_policy_register);
Vivek Goyal3e252062009-12-04 10:36:42 -05001139
Tejun Heo8bd435b2012-04-13 13:11:28 -07001140/**
Tejun Heo3c798392012-04-16 13:57:25 -07001141 * blkcg_policy_unregister - unregister a blkcg policy
1142 * @pol: blkcg policy to unregister
Tejun Heo8bd435b2012-04-13 13:11:28 -07001143 *
Tejun Heo3c798392012-04-16 13:57:25 -07001144 * Undo blkcg_policy_register(@pol). Might sleep.
Tejun Heo8bd435b2012-04-13 13:11:28 -07001145 */
Tejun Heo3c798392012-04-16 13:57:25 -07001146void blkcg_policy_unregister(struct blkcg_policy *pol)
Vivek Goyal3e252062009-12-04 10:36:42 -05001147{
Tejun Heobc0d6502012-04-13 13:11:26 -07001148 mutex_lock(&blkcg_pol_mutex);
1149
Tejun Heo3c798392012-04-16 13:57:25 -07001150 if (WARN_ON(blkcg_policy[pol->plid] != pol))
Tejun Heo8bd435b2012-04-13 13:11:28 -07001151 goto out_unlock;
1152
1153 /* kill the intf files first */
Tejun Heo3c798392012-04-16 13:57:25 -07001154 if (pol->cftypes)
1155 cgroup_rm_cftypes(&blkio_subsys, pol->cftypes);
Tejun Heo44ea53d2012-04-01 14:38:43 -07001156
Tejun Heo8bd435b2012-04-13 13:11:28 -07001157 /* unregister and update blkgs */
Tejun Heo3c798392012-04-16 13:57:25 -07001158 blkcg_policy[pol->plid] = NULL;
Tejun Heo8bd435b2012-04-13 13:11:28 -07001159out_unlock:
Tejun Heobc0d6502012-04-13 13:11:26 -07001160 mutex_unlock(&blkcg_pol_mutex);
Vivek Goyal3e252062009-12-04 10:36:42 -05001161}
Tejun Heo3c798392012-04-16 13:57:25 -07001162EXPORT_SYMBOL_GPL(blkcg_policy_unregister);